Jeopardy – IBM Challenge Extra Credit Paper Guidelines
Write at least a two page paper using Times New Roman 12 font double spaced that presents
your impressions of Watson’s performance on Jeopardy, the current state of Artificial
Intelligence (AI), what limitations Watson has, and future applications for the use of AI
technology. The paper should also answer the following questions:
a. Why is simply understanding the Jeopardy questions a difficulty for Watson?
Jeopardy questions are tricky because they contain jokes,
double meanings, and puns that are difficult to write traditional
rules based algorithms to solve. The subtleties of human
language are very difficult to translate into a finite set of rules
that a computer can understand. Human language is a mine field
for computers. Consider this sentence: I shot an elephant
wearing my pajamas. Who was wearing the pajamas? What is
meant by shot? If it was a photographer shooting it means
something very different then a hunter shooting. There is a lot of
ambiguity in natural language which is based on context and
computers have difficulty with context.
b. Why don’t rules-based algorithms work for programming artificial intelligence
computers?
Algorithms to solve problems that have well defined rules,
parameters, and goals like the game of chess are effectively
solved with rules-based algorithms, but natural language is
based on so many things that we as humans know or understand
that are difficult to capture into rules. There are just too many
common sense rules that would need to be defined to create
practical rules-based solutions to getting the computer to
understand natural language. One computer scientist has been
leading a team trying to define all the rules of knowledge for
human common sense and currently has 6 million rules that
make up about 3% of the total number of rules needed by the
Jeopardy – IBM Challenge Extra Credit Paper Guidelines
computer to understand human common sense. After 25 years of
efforts it is clear to most computer scientists that rule alone will
not be enough to build artificial intelligence.
c. What method of programming is used to improve on rule-based algorithms?
Watson uses a more powerful and flexible method to solve AI
problems called machine learning. Machine learning is simply
programming the computer to create its own rules by giving
millions of examples of a concept. In this way computers are
able to recognize patterns of the concept. Watson was trained to
recognize patterns using tens of thousands of old Jeopardy
questions along with their correct answers. Watson finds
patterns between the questions, answers, and the type of
evidence used to support those correct answers. Then weights
are given to the patterns to determine on average which solution
is most likely to be the correct answer to a brand new question
Watson has never heard before.
Computer learning plays a role in many modern computer
applications like weather prediction, recommendations from
Amazon or Netflix. Speech recognition is another example of
machine learning where thousands of examples of word
pronunciations are given to the computer to search for patterns
to make speech recognition more accurate.
Jeopardy – IBM Challenge Extra Credit Paper Guidelines
d. What final piece of information did Watson need to help it fully understand the
question categories on Jeopardy?
To boost Watson’s performance the correct answers to
questions solved by other contestants are feed as an electronic
text message back into the program. This final piece of feedback
helped Watson to understand the categories of questions and the
form of their answers by learning a new pattern on the fly.
Future applications of AI technology such as Watson could be a
computer that professionals such as medical doctors might
speak to verbally posing question that the system uses to search
huge databases of journals and emerging clinical research along
with patient history to help diagnose illnesses or recommend
treatments that the doctor may not even be up to date on
because of the fast pace of developments in the field.